KR20090089358A - A system and method for using a working global history register - Google Patents

Abstract

A method of processing branch history information is disclosed. The method retrieves branch instructions from an instruction cache and executes the branch instructions in a plurality of pipeline stages. The method verifies that a branch instruction has been identified. The method further receives branch history information during a first pipeline stage and loads the branch history information into a first register, wherein the first register. The method further loads the branch history information into the second register during the second pipeline stage. ® KIPO & WIPO 2009

Description

A SYSTEM AND METHOD FOR USING A WORKING GLOBAL HISTORY REGISTER}

The present invention relates generally to computer systems and, more particularly, to a method and system for using a working global history register.

Processors are the key to the evolution of computer platforms. Previous processors were limited by the technology available at that time. New advances in manufacturing technology allow transistor designs to be reduced to or exceed 1/1000 of the size of previous processors. These smaller processor designs are faster, more efficient, and use substantially lower power while providing processing power beyond previous expectations.

As the physical design of processors has evolved, so have innovative ways of processing information and performing functions. For example, "pipelining" of instructions has been implemented in processor designs since the early 1960s. One example of pipelining is the concept of breaking execution pipelines into units or stages, through which instructions flow sequentially in the stream. The stages are arranged so that several stages can process the appropriate portions of some instructions simultaneously. One advantage of pipelining is that the execution of instructions overlaps because the instructions are evolved in parallel.

The processor pipeline consists of many stages, each of which performs a function related to executing an instruction. Each stage is referred to as a pipe stage or pipe segment. The stages are connected together to form a pipeline. Instructions enter at one end of the pipeline and exit at the other end.

Most programs executed by a processor contain condition branch instructions, and the actual branch behavior of the instruction is not known until the instruction is deployed deep into the pipeline. In order to avoid stalls resulting from the actual deployment of branch instructions, modern processors may use some form of branch prediction, whereby branch behavior of conditional branch instructions is initially predicted in the pipeline. Based on the predicted branch evolution, the processor may execute instructions from the predicted address-either the branch target address (if the branch is expected to be taken) or the next sequential address after the branch instruction (if the branch is not expected to be taken). Fetch and execute speculatively. Whether a conditional branch instruction is taken or not taken is referred to as the determination of branch direction. The determination of the branch indication may be made at the prediction time and the actual branch resolution time. When the actual branch behavior was determined, if the branch was mispredicted, speculatively fetched instructions had to be flushed from the pipeline, and new instructions were fetched from the corrected address. Inferentially fetching instructions in response to bad branch prediction adversely affects processor performance and power consumption. In conclusion, improving the accuracy of branch predictions is an important processor design goal.

One known form of branch prediction involves partitioning the branch prediction into two predictors: the original branch target address cache (BTAC) and the branch history table (BHC). The BTAC is indexed by the instruction fetch group address and includes the following fetch address, corresponding to the instruction fetch group address, also referred to as the branch target. The entries are added to BTAC after the branch instruction passes through the processor pipeline, and its branch is taken. If BTAC is full, entries are removed from BTAC using standard cache replacement algorithms (such as round robin or last-recently used) when the next entry is added.

BTAC may be a highly-associative cache design and is initially accessed in the instruction execution pipeline. If the fetch group address matches a BTAC entry (BTAC hit), the corresponding next fetch address or target address is fetched in the next cycle. Subsequent fetching of this match and target address is referred to as branch prediction taken implicitly. If no match (BTAC miss) exists, the next sequentially incremented address is fetched in the next cycle. This no match situation is also referred to as a prediction that is not taken implicitly.

BTACs can be used with more accurate individual branch indication predictors, such as branch history tables (BHTs), also known as pattern history tables (PHTs). Conventional BHT may include a set of predicted indication counters that are saturated to produce more accurate taken / untaken decisions for individual branch instructions. For example, each saturated predicted indication counter may include a 2-bit counter that takes one of four states, each assigned a weighted prediction value as follows:

11-strong predictions are taken

10-weak predictions are taken

01-no weak predictions taken

00-No strong prediction is taken

The output of a conventional BHT, also referred to as a predicted value, does not make or make a decision to fetch the target address of a branch instruction or the address following the next cycle. The BHT is generally updated with branch result information as will be known.

To increase the accuracy of branch predictions, various other prediction techniques may be implemented that use recent branch history information from other branches as feedback. Those skilled in the art should recognize that current branch behavior is correlated with the history of previously executed branch instructions. For example, the history of previously executed branch instructions can affect how conditional branch instructions are predicted.

Global history registers (GHRs), also referred to in the art as global branch history registers or global history shift registers, can be used to track past history of previously executed branch instructions. As stored by the GHR, branch history provides a view of the sequence of branch instructions encountered in the code path leading to the branch instruction currently executed to achieve improved prediction results.

In some processors, the identification of the branch instruction and the prediction information related to it occur only after the instruction decoding stage. In general, the instruction decoding stage may be a later stage in the instruction execution sequence. After the instruction is decoded and confirmed by the branch instruction, the GHR is loaded with the appropriate branch history information. Since branch history information is identified, it is shifted to GHR. The output of the GHR is used to identify the prediction value stored in the BHT, which is used to predict the next condition branch instruction.

In a conventional processor using GHR, GHR does not represent the actual branch history information encountered when multiple branch instructions are executed in parallel for a relatively short period of time. In this example, the GHR is not updated with branch history information from the first branch instruction before the second branch instruction is predicted. As a result, an incorrect value of GHR can be used to identify an entry in the BHT for the second condition branch instruction. Using incorrect values to index entries in BHT can affect the accuracy of branch prediction. If the processor cannot follow the branch history information from the first condition branch instruction, another value is stored in the GHR and another entry of the BHT will be identified for the second condition branch instruction.

Thus, there is an industrial need to have a processor that can store and use branch history information rather than GHR to achieve more accurate branch predictions. The present invention discloses a processor that recognizes this need and identifies branch instructions at execution stages of the processor. Using the branch instruction information as input, the processor can steer the selection of prediction values for sequential condition branch instructions.

A method of processing branch history information is disclosed. The method identifies branch instructions during a first pipeline stage and loads the branch history information in a first register during the first pipeline stage. The method identifies the branch instructions at a second pipeline stage and the branch history information is loaded into a second register during the second pipeline stage.

A pipeline processor is disclosed that includes a first register containing branch history information and a second register containing branch history information. The pipeline processor includes a plurality of pipeline stages, wherein the first register is loaded with the branch history information at the first pipeline stage when a branch instruction is identified, and the second register has branch history during the second pipeline stage. The information is loaded.

A method of processing branch history information is disclosed. The method fetches a branch instruction, identifies the branch instructions during a first pipeline stage, and loads the branch history information in a first register during the first pipeline stage. The method confirms the branch instructions at a second pipeline stage, and the branch history information is loaded into a second register during the second pipeline stage.

In addition to further features and advantages of the present invention, a more complete understanding of the invention will be apparent from the following detailed description and the accompanying drawings.

1 illustrates a high level logic hardware block diagram of a processor using one embodiment of the present invention.

2 illustrates an example branch history table used by the processor of FIG. 1.

3 illustrates a low level logic block diagram of the processor of FIG. 1 using a working global history register.

4 shows a detailed view of the working global history register and the global history register.

5 illustrates an example group of instructions executed by the processor of FIG. 1.

FIG. 6 shows a timing diagram of the group of example instructions of FIG. 5 as executed through various stages of the processor of FIG. 1.

FIG. 7 shows a flow chart illustrating the instruction process flow performed by the processor of FIG. 1 using the working global history register.

DETAILED DESCRIPTION The detailed description is set forth below in connection with the accompanying drawings, which are intended as a description of various embodiments of the invention rather than to show the only embodiments in which the invention may be practiced. The detailed description includes specific details for the purpose of providing a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that the present invention may be practiced without these specific details. In some instances, well-known structures and components are shown in block diagram form in order to avoid obscuring the concepts of the present invention. Acronyms and other technical terms may be used merely for convenience and clarity and are not intended to limit the scope of the invention.

1 shows a high level diagram of a superscalar processor 100 using the embodiment described herein. The processor 100 has a central computing unit (CPU) 102 connected to the instruction cache 106 via a dedicated high speed bus 104. The instruction cache is also coupled to the memory 114 via the general purpose bus 116.

Within the processor 100, an instruction fetch unit (IFU) 122 controls the loading of instructions from the memory 114 to the instruction cache 106. Once the instruction cache 106 is loaded with instructions, the CPU 102 may access the instructions via the high speed bus 104. The instruction cache 106 may be a separate memory structure as shown in FIG. 1 or may be integrated as an internal component of the CPU 102. Integration may be based on the size of the instruction cache 106 as well as the complexity and power dissipation of the CPU 102. Branch target address cache (BTAC) 130, branch history table (BHT) 140 and two lower pipelines 160, 170 are also coupled to IFU 122.

The instructions may be fetched and decoded from the instruction cache 106, and several instructions may be fetched and decoded at one time. In the instruction cache 106, instructions are grouped into sections known as cache lines. Each cache line may contain a plurality of instructions as well as associated data. The number of instructions fetched may depend on the number of instructions in each cache line as well as the fetch bandwidth required. Within IFU 122, the fetched instructions are analyzed for operation type and data dependencies. After analyzing the instructions, processor 100 may distribute instructions from IFU 122 to lower functional units or lower pipelines 160 or 170 for further execution.

The lower pipelines 160, 170 may include various execution units EU 118 including computational logic units, floating point units, storage units, load units, and the like. For example, the EU 118, such as a computational logic unit, has a wide range of integer additions, subtractions, simple multiplications, bitwise logical operations (e.g. AND, NOT, OR, XOR), bit shifting, and the like. Calculation functions can be performed. Further, lower pipelines 160, 170 may have a resolution stage (not shown) while the actual results of the condition branch instruction are identified. Once the actual results of the branch instruction are identified, processor 100 can compare the actual results with the predicted results, and if they do not match, an incorrect prediction has occurred.

One of ordinary skill in the art may have BTAC 130 similar to branch target buffer (BTB) or branch target instruction cache (BTIC). The BTB or BTIC stores both the address of the branch and the instruction data (or opcodes) of the target branch. For convenience of description, BTAC 130 is used in connection with various embodiments of the present invention. Other embodiments of the invention may optionally include BTB or BTIC instead of BTAC 130.

At the first time the branch instruction is executed, there is no entry in BTAC 130 and a BTAC miss occurs. After a branch instruction terminates its execution, BTAC 130 may be updated sequentially to indicate the target address of the particular condition branch instruction as well as the processor mode (eg, Arm vs. Advanced RISC processor architecture). Thumb behavior). At any time a branch instruction is fetched again, the information stored in BTAC 130 will be fetched in the next processor cycle even if the branched instruction is not fully decoded.

A BTAC hit (eg, when a fetch group address matches an address in BTAC 130) may occur for a conditional or unconditional branch instruction. This is due to the fact that the BTAC 130 can store information relating to both conditional branch instructions as well as unconditional branch instructions. In the case of a BTAC hit for an unconditional branch instruction, not only the fact that the branch instruction is unconditional, but also the predicted target address, the predicted mode of the processor, can be stored. In situations where an unconditional branch instruction address is stored in the entry of BTAC 130, the entry will indicate the branch indication to be taken.

2 shows a more detailed description of an exemplary branch history table (BHT) 140 used by the processor 100. The BHT 140 may consist of 2 ^m lines 202 indexed using an address with m address bits. In one embodiment, nine bits of the address result in BHT 140 having 512 lines. Within each line 202, there are 2 ⁿ counters 204 when n is the number of bits used to select the appropriate counter. Also, three bits of the address can be used to select the counter 204, resulting in the BHT 140 with eight counters 204 per line 202. In one exemplary embodiment, fetch group address bits 12 through 4 may be used to select line 202 of BHT 140. Bits 3-1 of the fetch group address may be used to select a particular counter 204.

Processor 100 may identify branch instructions earlier in the instruction execution process prior to the instruction decoding stage. If branch instructions are identified earlier, branch history information, such as predicted values (conditional branch instructions) or branch instructions taken (unconditional branch instructions), may also be identified at the same time. The working global history register (WGHR) may be used by the processor 100 to receive and process branch history information earlier in the instruction execution processor. For example, the WGHR may store prediction values of conditional branch instructions as well as branch instructions of unconditional branch instructions. Optionally, the WGHR may only store prediction values of condition branch instructions. The output of the WGHR may be used to index the corresponding entry in the BHT 140 for the next condition branch instruction.

4 shows a low level logical block diagram 400 of a processor 100 that includes a working global history register (WGHR) 416. The upper pipe 450 is in the lower level block diagram 400. The top of the upper pipe is connected to the fetch logic circuit 402. The upper pipe 450 includes four instruction execution stages: an instruction cache 1 stage (IC1, 404), an instruction cache 2 stage (IC2, 406), an instruction data alignment stage (IDA, 408), and a decoding stage (DCD, 410). Include. It should be noted that pipe stages may be added or excluded from the upper pipe 450 without limiting the scope of the present invention. Upper Pipe 450, Working Global History Register (WGHR, 416), Global History Register (GHR, 414), Branch Correction Logic Circuits (BCL, 440), Select Mux 422, and Address Hashing Logic Circuits 420 In addition to the fetch logic circuit 402 may also be located within the IFU (122).

Processor 100 begins executing instructions, and fetch logic circuit 402 determines which instructions are fetched during IC1 stage 404. To retrieve the instructions, the fetch logic circuit 402 sends a fetch group address to the instruction cache 106. If a fetch group address is found in the instruction cache 106 (eg, instruction cache hit), the instructions are read from the hit cache line of the instruction cache 106 during the IC2 stage 406.

In parallel, during IC1 stage 404, processor 100 sends a fetch group address to BTAC 130. If processor 100 encounters a BTAC hit, the information stored in BTAC for the fetch group address is received during IC2 stage 406. As mentioned previously, the information stored in BTAC 130 may include branch information such as branch targets, processor mode, as well as branch instructions taken (in the case of unconditional branch instructions).

Also during IC1 stage 404, the fetch logic (circuit) sends the fetch group address to address hashing logic circuit 420. In the addressing hashing logic circuit 420, bits 12-4 of the fetch group address are computed with an exclusive OR (XOR'd) with the result of the selection mux 422. The result of the address hashing logic circuit 420 (eg, an XOR function) provides the address index to the BHT 140. As mentioned previously, bits 3-1 of the fetch group address can provide select bits to select the appropriate counter 204.

During IC2 stage 406, processor 100 reads the results from sending the instruction fetch group address to instruction cache 106, BTAC 130 and BHT 140. At IC2 stage 406, processor 100 determines if a BTAC hit occurs. When a BTAC hit is identified during IC2 stage 406, processor 100 also determines whether the branch is a conditional or unconditional branch instruction. Predicted values from the BHT 140 at IC2 stage 406 are also received and stored.

Because each cache line of the instruction cache 106 may include a plurality of instructions, individual instructions may need to be separated from the cache line. Similarly, data can be entangled with instructions in the cache line. The information from the cache line may need to be formatted and aligned in order to properly parse and execute the instructions. Sorting and formatting of instructions into individual executable instructions occur during IDA stage 408.

After the instructions are processed during the IDA stage 408, they are passed through the decoding (DCD) stage 410. During DCD stage 410, the instructions are analyzed to determine the type of instruction and what additional information or resources may be needed for further processing. Depending on the type of instruction or the current instruction load, the processor 100 may hold the instruction in the DCD stage 410, or the processor 100 may execute sub-pipelines 160 or 170) can pass the command. In DCD stage 410, processor 100 verifies the instruction with the condition branch instruction, and confirms the predicted value of the instruction (read during IC2 stage 406) from BHT 140. The accuracy of the predicted value will be checked during the stage of later instruction execution of either of the lower pipelines 160 or 170. Until the branch prediction is determined to be incorrect (eg, incorrect prediction), the processor 100 assumes that the prediction value is true and proceeds with fetching instructions based on this prediction.

The upper pipe 450 is connected to the working global history register (WGHR) 416. WGHR 416 allows the processor 100 to store and process branch history information related to branch instructions identified prior to DCD stage 410. In one embodiment, WGHR 416 may be loaded with predictions from BHT 140 for condition branch instructions when a BTAC hit occurs. As described above, a BTAC hit means that the instruction being fetched is a branch instruction and is associated with branch history information (eg, a predicted value for a conditional branch instruction or an indication taken for an unconditional branch instruction). Based on this condition, the processor 100 may use branch history information earlier for subsequent branch predictions (ie, branch history information is more likely) as opposed to waiting for a branch instruction to be confirmed during DCD stage 410. Recent). The output of the WGHR 414 is sent to the address hashing logic circuit 420 to determine the address index for the next entry of the BHT 140.

The timing at which branch history information becomes available depends on how quickly branch history information can be retrieved from BHT 140 and how quickly a BTAC hit can be acknowledged. In some processor designs, branch history information and BTAC hits may be received during IC2 stage 406. In other processor designs, branch history information and BTAC hits may be received during IDA stage 408. In other processor designs that incorporate other stages again with the stages described above, branch history information and BTAC hits may be available during these stages prior to the decoding stage.

In one embodiment, branch history information for condition branch instructions is shifted to WGHR 416 during IC2 stage 406 (when a BTAC hit occurs). In yet another embodiment, branch history information for both conditional branch instructions and unconditional branch instructions is shifted to the WGHR 416. In a further embodiment, WGHR 416 may be updated with branch history information during IDA stage 408. This situation may occur when the predicted value is stored in the BHT 140 or when the BTAC hit information is not available up to the IDA stage 408.

The selection mux 422 is configured to receive the output of the WGHR 416. In one embodiment, the output of the WGHR 416 is a nine bit value that includes the branch history of the last nine branch instructions processed by the processor 100. The result of the selection mux 422 is used as input to the address hashing logic circuit 420 indexed by the BHT 140 for the next condition branch instruction.

The GHR 414 operates very similarly to the WGHR 416 except that the GHR 414 can be loaded with branch history information during the DCD stage 410. The contents of the GHR 414 will mirror the contents of the WGHR 416 as the branch instruction passes through the DCD stage 410. Depending on the circumstances, the output of the GHR can be used to index the prediction value.

The output of the GHR 414 is connected to the selection mux 422. When a BTAC miss occurs and it is determined to be identified as a branch instruction taken during the DCD stage 410, the select mux 422 selects the output of the GHR 414 used by the address hashing logic circuit 420 for indexing. Instruct them to. In this example, GHR 414 is used because WGHR 416 has not yet had branch history information for the branch taken (due to a BTAC miss). Optionally, because the WGHR 416 can be updated by sequential fetched branch instructions prior to indexing the BHT 140 for the current branch instruction, the output of the GHR 414 also occurs when a BTAC miss occurs. Address hashing logic 420. In this example, the WGHR 414 does not represent an appropriate value for the current branch instruction, and if used by the address hashing logic circuit 420, an incorrect entry in the BHT 140 may be indexed.

The output of the GHR 414 is also coupled to the branch correction logic circuit BCL 440. If a false prediction occurs, the BCL 440 uses the GHR 414 to provide a "true" copy of the branch history information used for reconstruction purposes. If false prediction occurs, BCL 440 restores branch history information in both GHR 414 and WGHR 416. As mentioned above, when the branch instruction reaches the resolution stage and the actual results do not match the predicted results, wrong prediction occurs.

When false prediction occurs, the BCL 440 sends information to the fetch logic circuit 402 instructing the fetch logic circuit 402 to flush instructions that are fetched based on the mis-predicted condition branch instruction. To be more efficient, the BCL 440 may restore the GHR 414 and the WGHR 416 to the correct branch history information while providing the correct branch history information to the selection mux 422. When a false prediction occurs, processor 100 uses the output of BCL 440 (via selection mux 422) directed to address hashing logic circuit 420 for use in indexing the appropriate counter 204. You can choose.

When processor 100 encounters a wrong prediction, BCL 440 restores the GHR and WGHR to their appropriate values. In one embodiment, the BCL 440 may take a snapshot of the GHR 414 after the GHR 414 is loaded with the predicted value for the condition branch instruction. The BCL 440 may then invert the most recent prediction value (eg, MSB) of the GHR 414. By taking the opposite of the prediction value, the BCL 440 prepares a corrected value that should appear in the GHR 414 and the WGHR 416 if a wrong prediction occurs. For example, if after identifying the condition branch instruction and its predicted value during DCD stage 410, GHR 414 and BCL 440 are loaded with value "101011111" (MSB => LSB). The BCL 440 may flip to the MSB corresponding to the condition branch instruction and store the corrected value "001011111" linked to the condition branch instruction. Thus, if the condition branch instruction is incorrectly predicted, the corrected value is ready to be sent to the GHR 414, the WGHR 416, and the selection mux 422.

5 shows a detailed view 500 of the WGHR 416, GHR 414, and BCL 440. Within detailed diagram 500, WGHR selection mux 502 receives branch history information from IC2 stage 406, DCD stage 410 as well as corrected branch history information from BCL 440. The GHR selection mux 504 receives the branch history information from the DCD stage 410 and the corrected branch history information from the BCL 440.

The WGHR selection mux 502 selects which input is to be used to load branch history information into the WGHR 416. When false prediction occurs, the input from the BCL 440 takes precedence over the information sent from the IC2 stage 406 or the DCD stage 410. The BCL 440 takes precedence because branch history information following bad prediction may be associated with condition branch instructions fetched according to an incorrectly predicted branch path. Therefore, the branch history information passed by the IC2 stage or the DCD stage 410 may also be inaccurate.

If no false predictions occur, the input selection for the WGHR selection mux 502 may be determined according to the following examples, listed from highest priority to lowest priority:

a) If the branch instruction returns a BTAC miss during IC2 stage 406 but the prediction taken after it is decoded during DCD stage 410 ends, the branch history value checked during DCD stage 410 is determined by WGHR 416. Is shifted to The DCD stage 410 has priority in this case because it has been fetched after the predicted taken branch instruction that needs to be emitted. Therefore, any branch history information identified during IC2 stage 406 is discarded for subsequent branch instructions that may be ready to be written to WGHR 416 during the same processor cycle.

b) If DCD stage 410 does not execute the branch instruction associated with the BTAC miss, IC2 stage 406 will have the next highest priority. As long as the BTAC hit occurs for the branch instruction, the branch history information identified during IC2 stage 406 is shifted to WGHR 416.

c) If a branch instruction was previously identified as a BTAC hit and the associated branch history information is loaded according to example (b) already described, the WGHR 416 will be rewritten from the DCD stage 410 one or more times. . Of course, if the conditional branch instruction is a BTAC miss and the branch instruction is not predicted, then the WGHR 416 is written to this branch history information. Writing of WGHR 416 ensures that GHR 414 and WGHR 416 will be synchronized after the command passes through decoding stage 410.

GHR selection mux 504 selects the appropriate input used to update GHR 414. Similar to the WGHR selection logic 502, the GHR selection mux 504 gives the input from the BCL 440 the highest priority for the same reasons as described above. Thus, if false predictions do not occur, the GHR 414 is updated with branch history information identified during the DCD stage 410 for a particular branch instruction.

6 shows a timing diagram 600 of an exemplary group of instructions 500 moving through the upper pipeline 450. There are a plurality of branch instructions within an exemplary group of instructions 500. The X-axis 602 of FIG. 6 depicts a processor cycle and the Y-axis 604 is the stage of execution within the upper pipe 450 through which the instructions pass, as well as the contents of the GHR 414 and WGHR 416. Shows. The contents of the GHR 414 and WGHR 416 are written during one processor cycle and latched at the beginning of the next processor cycle. As shown in the dimming diagram, the latched contents of GHR 414 and WGHR 416 are shown. For convenience of description, only the three most significant bits of the GHR 414 and WGHR 416 are shown. When the instructions are executed, the instructions descend to the Y-axis 604.

In processor cycle 1, fetch logic circuit 402 sends a fetch group address to instruction cache 106, BTAC 130, and address hashing logic circuit 420 for instruction A. This is shown in the timing diagram 600 as the command A enters the IC1 stage 404. Also in processor cycle 1, the most significant three bits of GHR 414 and WGHR 416 are all zeros, indicating that all three executed last branch instructions have not been taken.

In processor cycle 2, the results of sending the fetch group address to the instruction cache 106, BTAC 130, and BHT 140 are received. This is shown in the timing diagram as the command A enters the IC2 stage 406. Since instruction cache 106 stores a plurality of instructions, instruction A + 4 is also shown to be retrieved along with instruction A of IC2 stage 406. Logic circuitry within IC2 stage 406 analyzes the information received from BTAC 130 and BHT 140. During IC2 stage 406, processor 100 determines that instruction A is a condition branch instruction (based on information from a BTAC hit) as well as the predicted value returned from BHT 140. In this example, command A is expected to be taken. The actual entry of BHT 140 for command A may be either strongly taken 11 or weakly taken 10. At the end of processor cycle 2, processor 100 loads " 1 " in the MSB of WGHR 416 to indicate the predicted value associated with condition branch instruction A. Since instruction A is to be predicted, the next sequential instruction A + 4 will not be the next instruction to be executed by instruction A + 4, so that instruction A may cause IC2 stage 406 to fail. Emitted after passing through. As shown in the timing diagram 600, the value “100” is latched into the WGHR 416 at the beginning of processor cycle three.

During processor cycle 3, instruction A enters IDA stage 408. While in IDA stage 408, command A is formatted and aligned, preparing the command to enter DCD stage 410. While instruction A moves through IDA stage 408, the fetch group address for instruction B is sent to instruction cache 106, BTAC 130, and BHT 140 during IC1 stage 404. .

In processor cycle 4, instruction A enters DCD stage 410, the results from the fetch request for instructions B and B + 4 are received, and the fetch group address for instruction B + 8 is Command cache 106, BTAC 130 and BHT 140 (IC1 stage 404). The contents ("100") of the WGHR 416 are selected by the selection mux 422 and used by the address hashing logic circuit 420 to index the entry into the BHT 140 for the instruction B + 8. do. When instruction A is in DCD stage 410, processor 100 confirms that instruction A is a condition branch instruction and as a result the predicted value "1" is shifted to GHR 414. The processor 100 will not know the updated value of the GHR 414 from the instruction A until the start of processor cycle 5 when the processor 100 latches the GHR 414. At the end of processor cycle 4, instruction A leaves upper pipe 450 and is directed to lower pipe 160 or 170 for further execution.

In a conventional processor using only GHR to store branch history information and an exemplary group of executed instructions 500 without using WGHR 416, the predicted returned from BHT for instruction B + 8 The value may not be accurate. This is because the address hashing logic will use the value of GHR in processor cycle 4 to determine the entry of the BHT for the instruction B + 8 (e.g., the value "000" was used). This value of GHR does not accurately represent the actual branch history encountered by the processor because the branch history information for instruction A did not appear correctly. If the same instruction sequence was executed sequentially, but this time the processor experienced a delay when fetching the instruction (B + 8) (ie, the address hashing logic would use the value of the GHR to access the BHT entry). Until the contents of the GHR have been updated), different entries into the BHT can be accessed. In such a case, a processor using only GHR to store branch history information may access two different BHT entries for the same condition branch instruction with the same instruction execution sequence.

In one embodiment, when the command A is in the DCD stage 410, the WGHR 416 is rewritten back to the predicted value at the same time that the GHR 414 is loaded. By writing both registers with the same predictive value at the same time, the two registers are synchronized for command A. Since it is rare to expect two condition branch instructions to be immediately taken following another instruction, the synchronization of the two registers is unlikely to lose any branch history information.

In processor cycle 5, instructions B + 8 and B + 12 enter IC2 stage 406 while instructions B and B + 4 enter IDA stage 408. Also in processor cycle 5, the fetch group address for instructions B + 16, B + 20 is sent to instruction cache 106, BTAC 130 and BHT 140. At IC2 stage 406, command B + 8 returns a BTAC hit. Since instruction B + 8 is a BTAC hit, processor 100 also determines that instruction B + 8 is a condition branch instruction, and its predicted value returned from BHT 140 during IC2 stage 406 is Shifted to WGHR 416. In this example, command B + 8 is also for which prediction is taken. The actual entry of the BHT 140 may be either strongly taken 11 or weakly taken 10. Since instruction B + 8 is a branch instruction for which prediction has been taken, instructions B + 12, B + 16, B + 20 cause instruction B + 8 to leave IC2 stage 406 (from a BTAC hit). The received) target address display command C will be issued by the fetch logic circuit 402 after being directed to the fetch logic circuit 402. The contents of WGHR 416 are updated with a predicted value of "taken" ("1"), which is latched at the beginning of processor cycle 6 as shown in timing diagram 600.

In processor cycle 6, instructions B and B + 4 enter DCD stage 410 while instruction B + 8 enters IDA stage 408. Also, during processor cycle 6, the fetch group address for instruction C is sent to instruction cache 106, BTAC 130, and BHT 140 (IC1 stage 404). At the end of processor cycle 6, instruction B, B + 4 leaves upper pipe 450 and is directed to lower pipelines 160 or 170 for further execution.

In processor cycle 7, instruction B + 8 is processed during DCD stage 410. During the DCD stage 410, the instruction B + 8 is verified with a condition branch instruction and its predicted value is also confirmed. The predicted value identified for instruction B + 8 is shifted to GHR 414 and reloaded to WGHR 416 during processor cycle 7. Instructions C and C + 4 are returned from the instruction cache 106 during the IC2 stage 406. At the end of processor cycle 7, instruction B + 8 leaves upper pipe 450 and is directed to lower pipelines 160 or 170 for further execution.

In a code segment in which branch instructions (based on the depth of the pipeline) can be executed in close proximity to each other, the most recent branch history information is used to process branch predictions.

During processor cycle 8, the value of GHR 414 is latched with WGHR 416. Instructions C and C + 4 may be processed during IDA stage 410 and any sequential instructions following instructions C and C + 4 may be fetched and executed.

7 is a flow chart illustrating an instruction process flow 700 taken by processor 100 to execute instructions using working global history register (WGHR) 416. Instruction process flow 700 begins at block 702. The instruction process flow proceeds to block 704 where the fetch logic circuit 402 sends the fetch group address to the BTAC 130 and the address hashing logic circuit 420 (to index into the BHT 140). As described above, the transfer of the fetch group address may occur at processor 100 during IC1 stage 404. At block 704, the results of the search of BTAC 130 (to determine whether the fetched instruction is a branch instruction) are returned. The results are returned during IC2 stage 406. From block 704, the instruction process flow 700 proceeds to decision block 706. Processor 100 determines whether a BTAC hit occurred at decision block 706. This determination may also occur during IC2 stage 406. As described above, a BTAC hit can occur for a conditional branch instruction or an unconditional branch instruction taken. If no BTAC hit exists (eg, a BTAC miss), the instruction processor flow 700 proceeds directly to block 712.

If there is a BTAC hit, the instruction process flow 700 proceeds to block 710. At block 710, the WGHR 416 is updated by shifting the prediction value retrieved from the BHT 140 to the WGHR 416. For example, if a branch instruction is taking predictions, " 1 " is shifted to WGHR 416, and if no prediction is taken, " 0 " is shifted. Depending on the implementation, the prediction value may be returned during any processor execution stage prior to the decoding stage. In the embodiment described above, the WGHR 416 is updated during the IC2 stage 406.

The instruction processor flow 700 proceeds to block 712 where the instruction passes through the decoding stage (eg, DCD stage 410). During the decoding stage, at block 712, the instruction can be identified as a branch instruction. After the instruction is executed at the decoding stage, the instruction processor flow 700 proceeds to decision block 714. If the instruction at decision block 714 is not a branch instruction, the instruction processor flow 700 ends at block 720.

If at block 714 the processor 100 confirms that the instruction is a branch instruction, the instruction process flow 700 proceeds to block 716. At block 716, the WGHR 416 and GHR 414 are updated with the appropriate branch history information, and the command process flow ends at block 720.

The various illustrative logic blocks, modules, circuits, components, and / or components described in connection with the embodiments described herein may be used in a general purpose processor, digital signal processor (DSP), application specific integrated circuit (ASIC), It may be implemented or performed through a field programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or a combination of those designed to implement these functions. A general purpose processor may be a microprocessor, but in alternative embodiments, such processor may be an existing processor, controller, microcontroller, or state machine. A processor may be implemented as a combination of computing devices, such as, for example, a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or a combination of these configurations.

Although specific embodiments have been shown and described herein, one of ordinary skill in the art can substitute for the specific embodiments in which any arrangement calculated to achieve the same purpose is shown, and the present invention provides for other applications in different environments. Be aware that you have Such application is intended to cover any changes or variations of the present invention. The following claims are not intended to limit the scope of the invention to the specific embodiments described herein.

Claims (22)

As a method of processing branch history information, Identifying a branch instruction at a first pipeline stage; Loading the branch history information into a first register during the first pipeline stage; Identifying the branch instruction at a second pipeline stage, The branch history information is loaded into a second register during the second pipeline stage. The method of claim 1, Identifying the branch instruction occurs when a branch target address cache (BTAC) hit is received. The method of claim 1, Identifying the branch instruction occurs when a branch target instruction cache (BTIC) hit is received. The method of claim 1, And wherein the first pipeline stage is an instruction cache stage. The method of claim 1, Wherein the first register and the second register are shift registers. The method of claim 5, Wherein the first register and the second register are 9-bit shift registers. The method of claim 1, Wherein the first and second registers store branch history information for conditional branch instructions. The method of claim 1, Wherein the first and second registers store branch history information for conditional branch instructions and unconditional branch instructions. The method of claim 1, And wherein said second pipeline stage is a decoding stage. As a pipeline processor, A first register containing branch history information, A second register containing branch history information, A plurality of pipeline stages, The branch register is loaded with the branch history information during a first pipeline stage when a branch instruction is identified, and the branch history information is loaded with a second register during a second pipeline stage. The method of claim 10, Wherein the branch instruction is identified when a branch target address cache (BTAC) hit occurs. The method of claim 10, The branch instruction is identified when a branch target instruction cache (BTIC) hit occurs. The method of claim 10, And the first pipeline stage is an instruction cache stage. The method of claim 10, The second pipeline stage is an instruction decoding stage. The method of claim 10, Wherein the branch history information further includes branch history information for conditional branch instructions. The method of claim 10, Wherein the branch history information further comprises branch history information for conditional branch instructions and unconditional branch instructions. The method of claim 10, Said first register and said second register are shift registers. The method of claim 10, And said second register is used to provide input to a branch correction logic circuit. As a method of processing branch history information, Fetching branch instructions; Identifying the branch instruction at a first pipeline stage; Loading the branch history information into a first register during the first pipeline stage; And Identifying the branch instruction at a second pipeline stage, And wherein the branch history information is loaded into a second register during the second pipeline stage. The method of claim 19, Identifying the branch instruction occurs when a branch target address cache (BTAC) hit is received. The method of claim 19, And wherein the first pipeline stage is an instruction cache stage. The method of claim 19, And wherein said second pipeline stage is a decoding stage.

Images